What's new in Orq AI
Every release, feature addition, and improvement. Most recent first.
CI release gating is now generally available
The CI integration that blocks merges when an eval metric regresses below its threshold is now out of beta. Install the Orq CI app on GitHub or GitLab, configure your threshold rules in the dashboard, and Orq reports a status check on every pull request. Works with GitHub Actions, GitLab CI, and any CI system that reads commit status checks. See the CI integration guide for setup instructions.
Span timeline view redesigned
The trace timeline now shows nested spans in a waterfall layout. You can see which LLM calls overlap, which tool calls block the critical path, and where the total latency is concentrated. Hover a span for prompt and completion preview without leaving the timeline. Filter by span type (llm, tool, handoff) using the controls above the chart.
TypeScript SDK now stable
The Orq TypeScript SDK exits beta with stable APIs. Works with Node.js 18+, Bun, and Deno. Auto-instruments calls to the OpenAI and Anthropic client libraries when you initialise the SDK before your agent starts. Full source map support for accurate stack traces. Install with npm install @orq/sdk or bun add @orq/sdk.
Fixed token count discrepancy for streaming completions
When using streaming completions from OpenAI or Anthropic, the token counts in Orq traces were showing the pre-stream estimate rather than the actual token usage from the final stream event. This is now fixed. Affected traces before this release will not be backfilled, but all new traces will show accurate counts.
LLM-as-judge evals and custom rubrics
You can now define LLM-as-judge eval metrics using a custom rubric prompt. Write your own scoring criteria, pick a judge model, and Orq runs the judge against every traced output. Results appear alongside deterministic scores in your eval dashboard. Score history per metric per version makes it straightforward to see if a prompt change improved or hurt quality on each rubric dimension.
Slack and webhook alerting
Connect Orq to a Slack channel or any webhook endpoint to receive alerts when a metric drifts outside its defined range. You can configure separate alert thresholds per metric: one for warning, one for critical. Orq sends one alert per triggered condition per hour to prevent notification fatigue. The alert message links directly to the trace that triggered it.
Stay up to date
We release improvements every two to three weeks. Follow along in the docs or get updates in our community channel.